search

geco-bern / agds

Applied Geodata Science book. Developed for the lecture(s) with the same name at the Institute of Geography, University of Bern.
https://geco-bern.github.io/agds/
Other
5 stars 6 forks source link

Inconsistency in explaining effect of hyperparameter tuning #73

Closed padasch closed 1 year ago

padasch commented 1 year ago

@stineb We discussed this before. As it is now, in MLII students see the result of hyperparameter tuning (poorer r2 on train but better r2 on test) but without additional explanation for why this is the case. This explanation is part of the report-exercise in MLI, so providing an in-depth explanation gives away the exercise.

But as it is written now in MLII, it reads like a loose end. Do you want to add some reference to the exercise or somehow state like "This result was to expected based on the theory given above".

https://github.com/geco-bern/agds/blob/81164435a8dcf0098174c90ba76c687a0873a5c5/10-supervised_ml_II.Rmd#L190

khufkens commented 1 year ago

This is still open, not addressed in the consistency check.

khufkens commented 1 year ago

@padasch @pepaaran, was this addressed in any follow up?

I would like to do a final reading tomorrow and then close the course materials (excluding examples solutions).

stineb commented 1 year ago

I'll read and comment/edit as I will read tomorrow and on Friday.